Best Bear Thesis on Nvidia?

Investment Analysis Clubs Saul’s Investing Discussions
1

Some/many board members have done well with NVDA, myself included. I often try to find reasons for investments that have done well for me to turn, and for Nvidia, the best I’ve come up with is:

Some Data Center build-outs are delayed or even cancelled. Data Center Watch counts $18 billion of projects blocked and another $46 billion delayed over roughly two years amid local opposition, with at least 142 activist groups across 24 states. Maybe that’s over-stated, as are some stats that half of all proposed Data Center build-outs have been cancelled, but there is some effect and some communities have put moratoriums in place for approvals. Power and water usage are often cited as concerns.

In Oct/Nov of last year, Microsoft flagged this as an issue, and even said that it is being forced to stockpile Nvidia GPUs since it can’t complete the data centers in time to deploy them. I believe this is why both Anthropic and Google were willing to pay higher than NeoCloud rates for instant, or in Google’s case, near-instant access to deployed GPUs from SpaceX.

When you factor in that the hyperscalers like Microsoft, Amazon, Google, and even Meta and Oracle are taking on debt to fund their AI build-outs, doesn’t it make sense that they would at least curtail future Nvidia purchases?

Now, there are some factors that work against this:

  1. Nvidia’s customer profile has shifted - hyperscalers are down to about 50% of the customer base, sovereign (government), NeoClouds, some industrial customers are on the rise.
  2. Hyperscalers don’t want to lose their place in line to buy Nvidia chips. Delaying previous orders means going to the back of what still is a long line and so they might have to wait beyond the Data Center build delay to get Nvidia servers.
  3. Overall demand is still very high, the order que is long, so even if some companies do drop out, it’s not like Nvidia is going to be sitting on product - they’ll probably still sell everything they make when they make them.

Anyone have thoughts or data on this?

27 Likes
2

I still have all of my shares purchased in 2008, 2009, 2010, and a little more in 2019 after the crypto mining downturn. I know some who sold after the crypto downturn in 2018, but I kept mine thinking about the future of self driving cars. Fast forward to today, for me, I firmly believe that these big companies will keep moving forward with build out in data centers for fear of falling behind in the race to be the best at AI. Despite the many activists and moratoriums, AI is going to move forward. There is no stopping it. The big companies will find a way around the obstacles - they always do. We all can think of the bad uses of AI, but hopefully the best use of AI may speed development of cures or advancement in many health issues like cancer, alzheimer’s, etc. Right now my investment in Nvidia has grown to 60% of my portfolios and I’m OK with that for now.

Razz

16 Likes
3

The biggest competitive threat I see is probably from the ASICs side of the business that nearly every hyperscaler is building out. Currently it looks like the hyperscalers are both buying NVIDIA machines and building there own.

To accomplish the same task way more ASICs are needed than a NVIDIA Blackwell though. However, these machines usually have less power requirements but take up more physical space to produce the same compute.

The limited availability of power and data centers play into this as well. If NVIDIA can prove out that their machines are just way more effective for any type of task it helps their business. If the ASIC machines can provide something comparable or good enough it hurts NVIDIA’s competitive positioning.

It’s worth keeping in mind that the market is so compute constrained that it seems like all sorts of chips and designs have demand right now, even if one machine performs better than another. It’s like AMD’s CEO said “it’s a huge market”. So even if AMD chips are underperforming NVIDIA’s, there is still huge demand for them.

Longer term NVIDIA’s designs have been so dominant for so long, but they are public from a high level of how they work. So while NVLink is proprietary, it may give ideas for another company to achieve the stronger performance with a new technology or architecture.

9 Likes
4

There certainly is a threat there, but Nvidia is already shippng a “Groq 3 LPX server” that pairs Vera-Rubins with LPUs. The LPX rack paired with Vera Rubin NVL72 delivers 35x higher inference throughput per megawatt than Blackwell NVL72 alone for trillion-parameter models, at a target of $45 per million tokens (from https://www.techi.com/nvidia-groq-deal-acqui-hire-ai-inference/ ).

This is from the Groq acquisition late last yer. Nvidia killed off its own prior “Rubin CPX” server effort as a result.

Funny enough, when I asked Claude about the ASIC threat to Nvidia, it gave me “real and accelerating,” yet “years away but directionally in motion.” When I asked Claude about Groq, it pushed back - hard - on me; when I asked about “Nvidia’s Groq chip,” saying there was no such thing. I had to tell Claude it was incorrect and cite the specific server name for Claude to apologize. So much for letting AI do your research. Claude Sonnet 4.6 was trained at data as of Aug 2025, so it missed the Dec 2025 acquisition of Groq. Something to keep in mind for the future in terms of using AI for research.

At any rate, the bottom line is, I believe:

The net result is that Nvidia’s revenue share will likely compress from ~87% at peak to somewhere in the 65-75% range by 2027, but absolute revenue continues to grow as the total market expands — and Nvidia now has a credible answer for both training and inference that it lacked a year ago.

It wouldn’t be surprising to see Nvidia’s uber-dominant share of AI server revenue decline from an amazing 87%, and 65%to 75% of a much larger pie still is a very good outlook (better than iPhone in its market, for instance).

That all said, @wpr101 is correct the ASICs are coming. The per chip ASIC’s cost about ⅓ of Nvidia’s GPUs (don’t know about LPUs), but the real factor is cost per token combined with token latency. For some/many inference applications, latency won’t matter much. For Agentic or Physical AI inference workloads, however, latency could be very important. Estimates are that LPUs are 10X-20X faster/lower latency. However, for “Amazon Basic” type build-outs (to steal a phrase from Dylan Patel), the cost for ASICs approaches an order of magnitude advantage - although as @wpr101 pointed out, that’s not factoring in data center infrastructure requiements (power, cooling, space, etc.).

Perhaps most telling is that Google, which has its own TPUs, decide to rent out SpaceX’s Nvidia GPUs at a high “surge pricing” cost. This is on top of Anthropic doing the same, even with Anthropic having Amazon with its Trainium and Inferentia ASICs available to it. Demand for compute remains very high, especially for almost instant availability. Does this moderate in say 3-5 years in a way that hurts Nvidia? Without knowing what Jensen has up his sleeves, I don’t think there’s a good way to tell. Mr. Market, however, is trading on the thinnest of AI narratives right now, so expect some volatility.

20 Likes
5

I don’t think inhibitions to data center build out or ASIC threats are the most important thing to consider with respect to Nvidia stock. Ultimately, we invest in companies because we want to realize stock price appreciation. All the research and analysis in which we heavily invest our time (our most precious and ever wasting asset) is directed at trying to best position our limited funds in order to produce the highest returns. I don’t think Nvidia is one of the best choices for meeting that goal despite their technical and market dominance in a high demand market which probably has legs for at least a couple more years.

The problem with Nvidia is market sentiment. It’s already the world’s largest company as measured by most (all?) of the common financial metrics of company size. Is that going to double in six months? Or a year? And then double again? Even if that were possible, a lot of people, myself included find it not just unlikely, but nearly inconceivable.

So whatever the merits of Rubin or Feynman or whatever physicist is chosen after that may be; and I’m confident, those merits will be mind boggling (can traces get any smaller than 3 microns? Can litho machines be built to make those at scale? - I guess the answer is yes, but at some point material science has got put a limit on what can be made). And Nvidia’s ability to maintain market dominance is virtually a certainty. And even the notion that once the build out is over, the market will crater is debatable. Plain old workhorse business computers were “refreshed” on about a 5 year cycle when I worked at Boeing. AI hardware won’t be any different, maybe a shorter refresh cycle.

It’s just to my way of thinking about investments, there are other really important considerations. And among them are what are the chances of making an early bet on a multi-bagger? If that’s something that’s important to you, Nvidia is probably not one of your top ten positions.

Full disclosure - I sold my position in Nvidia over a year ago.

8 Likes
6
  1. Reporting from The Information that Nvidia’s share of the Inference market is growing (from 66% last year to 74% today):
  1. [quote=“brittlerock, post:5, topic:125091”]
    It’s already the world’s largest company as measured by most (all?) of the common financial metrics of company size. Is that going to double in six months? Or a year? And then double again? Even if that were possible, a lot of people, myself included find it not just unlikely, but nearly inconceivable.
    [/quote]

• TTM revenue growth was “only” 71% as of the last ER. For the most recent quarter, it was up 85%YoY.
• Management forecasts $91 billion in total revenue next quarter, which would be 95% YoY growth.
• TTM EPS was 110.58%, so yeah, a literal double and then some in a year. For the most recent quarter, it was up 214%YoY.

Yet, NVDA is up only 43% in the TTM, slightly more if you include dividends.

Narrative-wise, there are still possible big wins from things like getting AI chips into computers and phones (like the Spark), as well as into physical AI, like robots. Where aren’t you going want inference?

I’m sure there are companies more likely to grow faster. And, truly aggressive investors may want their whole portfolio to contain a handful of such companies. But I think many growth investors will find in Nvidia high growth with stability that deserves a place in their portfolio.

27 Likes
7

@Smorgasbord1 I’m sure you’re right. I didn’t look up how many investors hold Nvidia. But, I’m sure it’s in the millions. Millions of investors hold Meta, Exxon, Apple, United Health and Pfizer and on and on and on. You could even add Bitcoin to the list. The SpaceX IPO valued the company just shy of the value placed on Microsoft, despite the non-existent financial record. Millions of people couldn’t wait to take a bite out of that one. Looks a lot like Westport to me. Arguing that lot’s of people hold stock in a company is hardly an argument for buying shares.

My point is, and actually, your post supports it, is that the stock price appreciation does not follow the expected pattern for a company with revenue growth and total financial profile that Nvidia displays. Most any smaller company with similar financial performance would display much more dramatic stock price appreciation. In fact, exactly like Nvidia displayed before it became a behemoth.

High growth investing has never been about buying stock in low risk companies. At least, I don’t recall Saul ever discussing the risk profile of a company. He held Amazon at one time but sold out long before it became the monster it has become.

The law of large numbers is a real thing. Even if Nvidia successfully defies that law, every investor who turns their back on the company today due to the sheer size of the company is not going to buy it tomorrow because it got bigger.

6 Likes
8

@brittlerock , we’re mostly on the same page with regards to NVDA, although I don’t understand where you got the idea that anything in my thesis is based on “how many investors hold Nvidia.”

My argument for Nvidia is two fold:

  1. The growth in the business’ performance every quarter continues to result in metrics (PE, PS, etc) that are undeniable, and so the stock price will appreciate accordingly. I agree the Law of Large Numbers is a real thing and also there’s plenty of doubt about Nvidia continuing to grow, even as it does quarter after quarter.
  2. Nvidia’s TAM is larger than just data centers, even with orbital data centers. It’s edge devices, including personal devices and eventually robotics.

Not everyone wants a totally aggressive growth portfolio of 8-12 small to mid size companies with high growth. A large cap growth company like Nvidia makes sense for some growth oriented investors. That’s all I was saying.

What happened there is that Amazon was ploughing almost all of its profits into its business infrastructure (eCommerce warehouses, logistics, and then data centers), so the P/E and other bottom-line metrics didn’t look good. However, Bezos was playing the long game and those investments eventually paid off big time. Saul’s method won’t capture those mid to long term investment wins; this isn’t a Buy And Hold board. Saul always said one can’t own every big winner, so missing AMZN didn’t bother him because what he did own also grew well.

Again, not sure why you’re bringing up “million of people” owning that in your response to me, as that’s not a factor in any of my arguments. But, with the danger of sidetracking into a SpaceX tangent, it appears you and many others will be surprised when SpaceX shows a profit this quarter and this fiscal year (depending on what Musk decides to spend on future data center build-outs). Analysts are pretty uniformly only looking at the past performance tables in the S1 and not at the S1 addendums of contracts with Anthropic and Google that are bringing in billions of dollars of revenue per month - at a soon to be $2.17B/month run-rate.

No doubt SPCX is going to be a rocky ride, and I do think the eventual TAM it put in the S1 is silly, but the company ain’t a Westport.

5 Likes
9

OK - So I guess I latched on to that sentence and over-reacted to it. Anyway, that’s where I got the idea that you considered the number of investors in a particular company a metric worth considering.

In any case, you said we’re mostly on the same page which I think is true, so I’ll stop trying to split hairs on minor disagreements.

As for SpaceX - I wouldn’t put a dime in that company for a host of reasons, not the least of which is that I find Musk to be a reprehensible individual. This is only relevant as an investor insofar as you think management is important. Musk is CEO and CoB for life. And any other title he awards himself, maybe Emperor. He alone will select the board members. He will select every member of the C-suite. Just not the governance structure of a company I want to own.

6 Likes
10

You’re still misreading it. It’s me saying Nvidia has enough growth to be a part of a Saul-style (“high growth”) portfolio for many people. It doesn’t come close to saying “invest because a lot of other people have invested.”

2 Likes
11

Uncle - I concede :slightly_smiling_face:

I wanted to keep this short. The editor will not publish a short message. This is filler.

5 Likes